xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision 0c6dfa75)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
4  *  Copyright 2007-2010 Freescale Semiconductor, Inc.
5  *
6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
7  *  and Paul Mackerras (paulus@samba.org)
8  */
9 
10 /*
11  * This file handles the architecture-dependent parts of hardware exceptions
12  */
13 
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/pkeys.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/ptrace.h>
23 #include <linux/user.h>
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/extable.h>
27 #include <linux/module.h>	/* print_modules */
28 #include <linux/prctl.h>
29 #include <linux/delay.h>
30 #include <linux/kprobes.h>
31 #include <linux/kexec.h>
32 #include <linux/backlight.h>
33 #include <linux/bug.h>
34 #include <linux/kdebug.h>
35 #include <linux/ratelimit.h>
36 #include <linux/context_tracking.h>
37 #include <linux/smp.h>
38 #include <linux/console.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/debugfs.h>
41 
42 #include <asm/emulated_ops.h>
43 #include <linux/uaccess.h>
44 #include <asm/interrupt.h>
45 #include <asm/io.h>
46 #include <asm/machdep.h>
47 #include <asm/rtas.h>
48 #include <asm/pmc.h>
49 #include <asm/reg.h>
50 #ifdef CONFIG_PMAC_BACKLIGHT
51 #include <asm/backlight.h>
52 #endif
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/processor.h>
56 #endif
57 #include <asm/kexec.h>
58 #include <asm/ppc-opcode.h>
59 #include <asm/rio.h>
60 #include <asm/fadump.h>
61 #include <asm/switch_to.h>
62 #include <asm/tm.h>
63 #include <asm/debug.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/hmi.h>
66 #include <sysdev/fsl_pci.h>
67 #include <asm/kprobes.h>
68 #include <asm/stacktrace.h>
69 #include <asm/nmi.h>
70 #include <asm/disassemble.h>
71 #include <asm/udbg.h>
72 
73 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
74 int (*__debugger)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
80 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
81 
82 EXPORT_SYMBOL(__debugger);
83 EXPORT_SYMBOL(__debugger_ipi);
84 EXPORT_SYMBOL(__debugger_bpt);
85 EXPORT_SYMBOL(__debugger_sstep);
86 EXPORT_SYMBOL(__debugger_iabr_match);
87 EXPORT_SYMBOL(__debugger_break_match);
88 EXPORT_SYMBOL(__debugger_fault_handler);
89 #endif
90 
91 /* Transactional Memory trap debug */
92 #ifdef TM_DEBUG_SW
93 #define TM_DEBUG(x...) printk(KERN_INFO x)
94 #else
95 #define TM_DEBUG(x...) do { } while(0)
96 #endif
97 
98 static const char *signame(int signr)
99 {
100 	switch (signr) {
101 	case SIGBUS:	return "bus error";
102 	case SIGFPE:	return "floating point exception";
103 	case SIGILL:	return "illegal instruction";
104 	case SIGSEGV:	return "segfault";
105 	case SIGTRAP:	return "unhandled trap";
106 	}
107 
108 	return "unknown signal";
109 }
110 
111 /*
112  * Trap & Exception support
113  */
114 
115 #ifdef CONFIG_PMAC_BACKLIGHT
116 static void pmac_backlight_unblank(void)
117 {
118 	mutex_lock(&pmac_backlight_mutex);
119 	if (pmac_backlight) {
120 		struct backlight_properties *props;
121 
122 		props = &pmac_backlight->props;
123 		props->brightness = props->max_brightness;
124 		props->power = FB_BLANK_UNBLANK;
125 		backlight_update_status(pmac_backlight);
126 	}
127 	mutex_unlock(&pmac_backlight_mutex);
128 }
129 #else
130 static inline void pmac_backlight_unblank(void) { }
131 #endif
132 
133 /*
134  * If oops/die is expected to crash the machine, return true here.
135  *
136  * This should not be expected to be 100% accurate, there may be
137  * notifiers registered or other unexpected conditions that may bring
138  * down the kernel. Or if the current process in the kernel is holding
139  * locks or has other critical state, the kernel may become effectively
140  * unusable anyway.
141  */
142 bool die_will_crash(void)
143 {
144 	if (should_fadump_crash())
145 		return true;
146 	if (kexec_should_crash(current))
147 		return true;
148 	if (in_interrupt() || panic_on_oops ||
149 			!current->pid || is_global_init(current))
150 		return true;
151 
152 	return false;
153 }
154 
155 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
156 static int die_owner = -1;
157 static unsigned int die_nest_count;
158 static int die_counter;
159 
160 extern void panic_flush_kmsg_start(void)
161 {
162 	/*
163 	 * These are mostly taken from kernel/panic.c, but tries to do
164 	 * relatively minimal work. Don't use delay functions (TB may
165 	 * be broken), don't crash dump (need to set a firmware log),
166 	 * don't run notifiers. We do want to get some information to
167 	 * Linux console.
168 	 */
169 	console_verbose();
170 	bust_spinlocks(1);
171 }
172 
173 extern void panic_flush_kmsg_end(void)
174 {
175 	kmsg_dump(KMSG_DUMP_PANIC);
176 	bust_spinlocks(0);
177 	debug_locks_off();
178 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
179 }
180 
181 static unsigned long oops_begin(struct pt_regs *regs)
182 {
183 	int cpu;
184 	unsigned long flags;
185 
186 	oops_enter();
187 
188 	/* racy, but better than risking deadlock. */
189 	raw_local_irq_save(flags);
190 	cpu = smp_processor_id();
191 	if (!arch_spin_trylock(&die_lock)) {
192 		if (cpu == die_owner)
193 			/* nested oops. should stop eventually */;
194 		else
195 			arch_spin_lock(&die_lock);
196 	}
197 	die_nest_count++;
198 	die_owner = cpu;
199 	console_verbose();
200 	bust_spinlocks(1);
201 	if (machine_is(powermac))
202 		pmac_backlight_unblank();
203 	return flags;
204 }
205 NOKPROBE_SYMBOL(oops_begin);
206 
207 static void oops_end(unsigned long flags, struct pt_regs *regs,
208 			       int signr)
209 {
210 	bust_spinlocks(0);
211 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
212 	die_nest_count--;
213 	oops_exit();
214 	printk("\n");
215 	if (!die_nest_count) {
216 		/* Nest count reaches zero, release the lock. */
217 		die_owner = -1;
218 		arch_spin_unlock(&die_lock);
219 	}
220 	raw_local_irq_restore(flags);
221 
222 	/*
223 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
224 	 */
225 	if (TRAP(regs) == INTERRUPT_SYSTEM_RESET)
226 		return;
227 
228 	crash_fadump(regs, "die oops");
229 
230 	if (kexec_should_crash(current))
231 		crash_kexec(regs);
232 
233 	if (!signr)
234 		return;
235 
236 	/*
237 	 * While our oops output is serialised by a spinlock, output
238 	 * from panic() called below can race and corrupt it. If we
239 	 * know we are going to panic, delay for 1 second so we have a
240 	 * chance to get clean backtraces from all CPUs that are oopsing.
241 	 */
242 	if (in_interrupt() || panic_on_oops || !current->pid ||
243 	    is_global_init(current)) {
244 		mdelay(MSEC_PER_SEC);
245 	}
246 
247 	if (panic_on_oops)
248 		panic("Fatal exception");
249 	make_task_dead(signr);
250 }
251 NOKPROBE_SYMBOL(oops_end);
252 
253 static char *get_mmu_str(void)
254 {
255 	if (early_radix_enabled())
256 		return " MMU=Radix";
257 	if (early_mmu_has_feature(MMU_FTR_HPTE_TABLE))
258 		return " MMU=Hash";
259 	return "";
260 }
261 
262 static int __die(const char *str, struct pt_regs *regs, long err)
263 {
264 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
265 
266 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s %s\n",
267 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
268 	       PAGE_SIZE / 1024, get_mmu_str(),
269 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
270 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
271 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
272 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
273 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
274 	       ppc_md.name ? ppc_md.name : "");
275 
276 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
277 		return 1;
278 
279 	print_modules();
280 	show_regs(regs);
281 
282 	return 0;
283 }
284 NOKPROBE_SYMBOL(__die);
285 
286 void die(const char *str, struct pt_regs *regs, long err)
287 {
288 	unsigned long flags;
289 
290 	/*
291 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
292 	 */
293 	if (TRAP(regs) != INTERRUPT_SYSTEM_RESET) {
294 		if (debugger(regs))
295 			return;
296 	}
297 
298 	flags = oops_begin(regs);
299 	if (__die(str, regs, err))
300 		err = 0;
301 	oops_end(flags, regs, err);
302 }
303 NOKPROBE_SYMBOL(die);
304 
305 void user_single_step_report(struct pt_regs *regs)
306 {
307 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
308 }
309 
310 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
311 			    unsigned long addr)
312 {
313 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
314 				      DEFAULT_RATELIMIT_BURST);
315 
316 	if (!show_unhandled_signals)
317 		return;
318 
319 	if (!unhandled_signal(current, signr))
320 		return;
321 
322 	if (!__ratelimit(&rs))
323 		return;
324 
325 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
326 		current->comm, current->pid, signame(signr), signr,
327 		addr, regs->nip, regs->link, code);
328 
329 	print_vma_addr(KERN_CONT " in ", regs->nip);
330 
331 	pr_cont("\n");
332 
333 	show_user_instructions(regs);
334 }
335 
336 static bool exception_common(int signr, struct pt_regs *regs, int code,
337 			      unsigned long addr)
338 {
339 	if (!user_mode(regs)) {
340 		die("Exception in kernel mode", regs, signr);
341 		return false;
342 	}
343 
344 	/*
345 	 * Must not enable interrupts even for user-mode exception, because
346 	 * this can be called from machine check, which may be a NMI or IRQ
347 	 * which don't like interrupts being enabled. Could check for
348 	 * in_hardirq || in_nmi perhaps, but there doesn't seem to be a good
349 	 * reason why _exception() should enable irqs for an exception handler,
350 	 * the handlers themselves do that directly.
351 	 */
352 
353 	show_signal_msg(signr, regs, code, addr);
354 
355 	current->thread.trap_nr = code;
356 
357 	return true;
358 }
359 
360 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
361 {
362 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
363 		return;
364 
365 	force_sig_pkuerr((void __user *) addr, key);
366 }
367 
368 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
369 {
370 	if (!exception_common(signr, regs, code, addr))
371 		return;
372 
373 	force_sig_fault(signr, code, (void __user *)addr);
374 }
375 
376 /*
377  * The interrupt architecture has a quirk in that the HV interrupts excluding
378  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
379  * that an interrupt handler must do is save off a GPR into a scratch register,
380  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
381  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
382  * that it is non-reentrant, which leads to random data corruption.
383  *
384  * The solution is for NMI interrupts in HV mode to check if they originated
385  * from these critical HV interrupt regions. If so, then mark them not
386  * recoverable.
387  *
388  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
389  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
390  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
391  * that would work. However any other guest OS that may have the SPRG live
392  * and MSR[RI]=1 could encounter silent corruption.
393  *
394  * Builds that do not support KVM could take this second option to increase
395  * the recoverability of NMIs.
396  */
397 noinstr void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
398 {
399 #ifdef CONFIG_PPC_POWERNV
400 	unsigned long kbase = (unsigned long)_stext;
401 	unsigned long nip = regs->nip;
402 
403 	if (!(regs->msr & MSR_RI))
404 		return;
405 	if (!(regs->msr & MSR_HV))
406 		return;
407 	if (regs->msr & MSR_PR)
408 		return;
409 
410 	/*
411 	 * Now test if the interrupt has hit a range that may be using
412 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
413 	 * problem ranges all run un-relocated. Test real and virt modes
414 	 * at the same time by dropping the high bit of the nip (virt mode
415 	 * entry points still have the +0x4000 offset).
416 	 */
417 	nip &= ~0xc000000000000000ULL;
418 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
419 		goto nonrecoverable;
420 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
421 		goto nonrecoverable;
422 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
423 		goto nonrecoverable;
424 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
425 		goto nonrecoverable;
426 
427 	/* Trampoline code runs un-relocated so subtract kbase. */
428 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
429 			nip < (unsigned long)(end_real_trampolines - kbase))
430 		goto nonrecoverable;
431 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
432 			nip < (unsigned long)(end_virt_trampolines - kbase))
433 		goto nonrecoverable;
434 	return;
435 
436 nonrecoverable:
437 	regs->msr &= ~MSR_RI;
438 	local_paca->hsrr_valid = 0;
439 	local_paca->srr_valid = 0;
440 #endif
441 }
442 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception)
443 {
444 	unsigned long hsrr0, hsrr1;
445 	bool saved_hsrrs = false;
446 
447 	/*
448 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
449 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
450 	 * OPAL), so save them here and restore them before returning.
451 	 *
452 	 * Machine checks don't need to save HSRRs, as the real mode handler
453 	 * is careful to avoid them, and the regular handler is not delivered
454 	 * as an NMI.
455 	 */
456 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
457 		hsrr0 = mfspr(SPRN_HSRR0);
458 		hsrr1 = mfspr(SPRN_HSRR1);
459 		saved_hsrrs = true;
460 	}
461 
462 	hv_nmi_check_nonrecoverable(regs);
463 
464 	__this_cpu_inc(irq_stat.sreset_irqs);
465 
466 	/* See if any machine dependent calls */
467 	if (ppc_md.system_reset_exception) {
468 		if (ppc_md.system_reset_exception(regs))
469 			goto out;
470 	}
471 
472 	if (debugger(regs))
473 		goto out;
474 
475 	kmsg_dump(KMSG_DUMP_OOPS);
476 	/*
477 	 * A system reset is a request to dump, so we always send
478 	 * it through the crashdump code (if fadump or kdump are
479 	 * registered).
480 	 */
481 	crash_fadump(regs, "System Reset");
482 
483 	crash_kexec(regs);
484 
485 	/*
486 	 * We aren't the primary crash CPU. We need to send it
487 	 * to a holding pattern to avoid it ending up in the panic
488 	 * code.
489 	 */
490 	crash_kexec_secondary(regs);
491 
492 	/*
493 	 * No debugger or crash dump registered, print logs then
494 	 * panic.
495 	 */
496 	die("System Reset", regs, SIGABRT);
497 
498 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
499 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
500 	nmi_panic(regs, "System Reset");
501 
502 out:
503 #ifdef CONFIG_PPC_BOOK3S_64
504 	BUG_ON(get_paca()->in_nmi == 0);
505 	if (get_paca()->in_nmi > 1)
506 		die("Unrecoverable nested System Reset", regs, SIGABRT);
507 #endif
508 	/* Must die if the interrupt is not recoverable */
509 	if (regs_is_unrecoverable(regs)) {
510 		/* For the reason explained in die_mce, nmi_exit before die */
511 		nmi_exit();
512 		die("Unrecoverable System Reset", regs, SIGABRT);
513 	}
514 
515 	if (saved_hsrrs) {
516 		mtspr(SPRN_HSRR0, hsrr0);
517 		mtspr(SPRN_HSRR1, hsrr1);
518 	}
519 
520 	/* What should we do here? We could issue a shutdown or hard reset. */
521 
522 	return 0;
523 }
524 
525 /*
526  * I/O accesses can cause machine checks on powermacs.
527  * Check if the NIP corresponds to the address of a sync
528  * instruction for which there is an entry in the exception
529  * table.
530  *  -- paulus.
531  */
532 static inline int check_io_access(struct pt_regs *regs)
533 {
534 #ifdef CONFIG_PPC32
535 	unsigned long msr = regs->msr;
536 	const struct exception_table_entry *entry;
537 	unsigned int *nip = (unsigned int *)regs->nip;
538 
539 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
540 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
541 		/*
542 		 * Check that it's a sync instruction, or somewhere
543 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
544 		 * As the address is in the exception table
545 		 * we should be able to read the instr there.
546 		 * For the debug message, we look at the preceding
547 		 * load or store.
548 		 */
549 		if (*nip == PPC_RAW_NOP())
550 			nip -= 2;
551 		else if (*nip == PPC_RAW_ISYNC())
552 			--nip;
553 		if (*nip == PPC_RAW_SYNC() || get_op(*nip) == OP_TRAP) {
554 			unsigned int rb;
555 
556 			--nip;
557 			rb = (*nip >> 11) & 0x1f;
558 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
559 			       (*nip & 0x100)? "OUT to": "IN from",
560 			       regs->gpr[rb] - _IO_BASE, nip);
561 			regs_set_recoverable(regs);
562 			regs_set_return_ip(regs, extable_fixup(entry));
563 			return 1;
564 		}
565 	}
566 #endif /* CONFIG_PPC32 */
567 	return 0;
568 }
569 
570 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
571 /* On 4xx, the reason for the machine check or program exception
572    is in the ESR. */
573 #define get_reason(regs)	((regs)->esr)
574 #define REASON_FP		ESR_FP
575 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
576 #define REASON_PRIVILEGED	ESR_PPR
577 #define REASON_TRAP		ESR_PTR
578 #define REASON_PREFIXED		0
579 #define REASON_BOUNDARY		0
580 
581 /* single-step stuff */
582 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
583 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
584 #define clear_br_trace(regs)	do {} while(0)
585 #else
586 /* On non-4xx, the reason for the machine check or program
587    exception is in the MSR. */
588 #define get_reason(regs)	((regs)->msr)
589 #define REASON_TM		SRR1_PROGTM
590 #define REASON_FP		SRR1_PROGFPE
591 #define REASON_ILLEGAL		SRR1_PROGILL
592 #define REASON_PRIVILEGED	SRR1_PROGPRIV
593 #define REASON_TRAP		SRR1_PROGTRAP
594 #define REASON_PREFIXED		SRR1_PREFIXED
595 #define REASON_BOUNDARY		SRR1_BOUNDARY
596 
597 #define single_stepping(regs)	((regs)->msr & MSR_SE)
598 #define clear_single_step(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_SE))
599 #define clear_br_trace(regs)	(regs_set_return_msr((regs), (regs)->msr & ~MSR_BE))
600 #endif
601 
602 #define inst_length(reason)	(((reason) & REASON_PREFIXED) ? 8 : 4)
603 
604 #if defined(CONFIG_PPC_E500)
605 int machine_check_e500mc(struct pt_regs *regs)
606 {
607 	unsigned long mcsr = mfspr(SPRN_MCSR);
608 	unsigned long pvr = mfspr(SPRN_PVR);
609 	unsigned long reason = mcsr;
610 	int recoverable = 1;
611 
612 	if (reason & MCSR_LD) {
613 		recoverable = fsl_rio_mcheck_exception(regs);
614 		if (recoverable == 1)
615 			goto silent_out;
616 	}
617 
618 	printk("Machine check in kernel mode.\n");
619 	printk("Caused by (from MCSR=%lx): ", reason);
620 
621 	if (reason & MCSR_MCP)
622 		pr_cont("Machine Check Signal\n");
623 
624 	if (reason & MCSR_ICPERR) {
625 		pr_cont("Instruction Cache Parity Error\n");
626 
627 		/*
628 		 * This is recoverable by invalidating the i-cache.
629 		 */
630 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
631 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
632 			;
633 
634 		/*
635 		 * This will generally be accompanied by an instruction
636 		 * fetch error report -- only treat MCSR_IF as fatal
637 		 * if it wasn't due to an L1 parity error.
638 		 */
639 		reason &= ~MCSR_IF;
640 	}
641 
642 	if (reason & MCSR_DCPERR_MC) {
643 		pr_cont("Data Cache Parity Error\n");
644 
645 		/*
646 		 * In write shadow mode we auto-recover from the error, but it
647 		 * may still get logged and cause a machine check.  We should
648 		 * only treat the non-write shadow case as non-recoverable.
649 		 */
650 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
651 		 * is not implemented but L1 data cache always runs in write
652 		 * shadow mode. Hence on data cache parity errors HW will
653 		 * automatically invalidate the L1 Data Cache.
654 		 */
655 		if (PVR_VER(pvr) != PVR_VER_E6500) {
656 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
657 				recoverable = 0;
658 		}
659 	}
660 
661 	if (reason & MCSR_L2MMU_MHIT) {
662 		pr_cont("Hit on multiple TLB entries\n");
663 		recoverable = 0;
664 	}
665 
666 	if (reason & MCSR_NMI)
667 		pr_cont("Non-maskable interrupt\n");
668 
669 	if (reason & MCSR_IF) {
670 		pr_cont("Instruction Fetch Error Report\n");
671 		recoverable = 0;
672 	}
673 
674 	if (reason & MCSR_LD) {
675 		pr_cont("Load Error Report\n");
676 		recoverable = 0;
677 	}
678 
679 	if (reason & MCSR_ST) {
680 		pr_cont("Store Error Report\n");
681 		recoverable = 0;
682 	}
683 
684 	if (reason & MCSR_LDG) {
685 		pr_cont("Guarded Load Error Report\n");
686 		recoverable = 0;
687 	}
688 
689 	if (reason & MCSR_TLBSYNC)
690 		pr_cont("Simultaneous tlbsync operations\n");
691 
692 	if (reason & MCSR_BSL2_ERR) {
693 		pr_cont("Level 2 Cache Error\n");
694 		recoverable = 0;
695 	}
696 
697 	if (reason & MCSR_MAV) {
698 		u64 addr;
699 
700 		addr = mfspr(SPRN_MCAR);
701 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
702 
703 		pr_cont("Machine Check %s Address: %#llx\n",
704 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
705 	}
706 
707 silent_out:
708 	mtspr(SPRN_MCSR, mcsr);
709 	return mfspr(SPRN_MCSR) == 0 && recoverable;
710 }
711 
712 int machine_check_e500(struct pt_regs *regs)
713 {
714 	unsigned long reason = mfspr(SPRN_MCSR);
715 
716 	if (reason & MCSR_BUS_RBERR) {
717 		if (fsl_rio_mcheck_exception(regs))
718 			return 1;
719 		if (fsl_pci_mcheck_exception(regs))
720 			return 1;
721 	}
722 
723 	printk("Machine check in kernel mode.\n");
724 	printk("Caused by (from MCSR=%lx): ", reason);
725 
726 	if (reason & MCSR_MCP)
727 		pr_cont("Machine Check Signal\n");
728 	if (reason & MCSR_ICPERR)
729 		pr_cont("Instruction Cache Parity Error\n");
730 	if (reason & MCSR_DCP_PERR)
731 		pr_cont("Data Cache Push Parity Error\n");
732 	if (reason & MCSR_DCPERR)
733 		pr_cont("Data Cache Parity Error\n");
734 	if (reason & MCSR_BUS_IAERR)
735 		pr_cont("Bus - Instruction Address Error\n");
736 	if (reason & MCSR_BUS_RAERR)
737 		pr_cont("Bus - Read Address Error\n");
738 	if (reason & MCSR_BUS_WAERR)
739 		pr_cont("Bus - Write Address Error\n");
740 	if (reason & MCSR_BUS_IBERR)
741 		pr_cont("Bus - Instruction Data Error\n");
742 	if (reason & MCSR_BUS_RBERR)
743 		pr_cont("Bus - Read Data Bus Error\n");
744 	if (reason & MCSR_BUS_WBERR)
745 		pr_cont("Bus - Write Data Bus Error\n");
746 	if (reason & MCSR_BUS_IPERR)
747 		pr_cont("Bus - Instruction Parity Error\n");
748 	if (reason & MCSR_BUS_RPERR)
749 		pr_cont("Bus - Read Parity Error\n");
750 
751 	return 0;
752 }
753 
754 int machine_check_generic(struct pt_regs *regs)
755 {
756 	return 0;
757 }
758 #elif defined(CONFIG_PPC32)
759 int machine_check_generic(struct pt_regs *regs)
760 {
761 	unsigned long reason = regs->msr;
762 
763 	printk("Machine check in kernel mode.\n");
764 	printk("Caused by (from SRR1=%lx): ", reason);
765 	switch (reason & 0x601F0000) {
766 	case 0x80000:
767 		pr_cont("Machine check signal\n");
768 		break;
769 	case 0x40000:
770 	case 0x140000:	/* 7450 MSS error and TEA */
771 		pr_cont("Transfer error ack signal\n");
772 		break;
773 	case 0x20000:
774 		pr_cont("Data parity error signal\n");
775 		break;
776 	case 0x10000:
777 		pr_cont("Address parity error signal\n");
778 		break;
779 	case 0x20000000:
780 		pr_cont("L1 Data Cache error\n");
781 		break;
782 	case 0x40000000:
783 		pr_cont("L1 Instruction Cache error\n");
784 		break;
785 	case 0x00100000:
786 		pr_cont("L2 data cache parity error\n");
787 		break;
788 	default:
789 		pr_cont("Unknown values in msr\n");
790 	}
791 	return 0;
792 }
793 #endif /* everything else */
794 
795 void die_mce(const char *str, struct pt_regs *regs, long err)
796 {
797 	/*
798 	 * The machine check wants to kill the interrupted context,
799 	 * but make_task_dead() checks for in_interrupt() and panics
800 	 * in that case, so exit the irq/nmi before calling die.
801 	 */
802 	if (in_nmi())
803 		nmi_exit();
804 	else
805 		irq_exit();
806 	die(str, regs, err);
807 }
808 
809 /*
810  * BOOK3S_64 does not usually call this handler as a non-maskable interrupt
811  * (it uses its own early real-mode handler to handle the MCE proper
812  * and then raises irq_work to call this handler when interrupts are
813  * enabled). The only time when this is not true is if the early handler
814  * is unrecoverable, then it does call this directly to try to get a
815  * message out.
816  */
817 static void __machine_check_exception(struct pt_regs *regs)
818 {
819 	int recover = 0;
820 
821 	__this_cpu_inc(irq_stat.mce_exceptions);
822 
823 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
824 
825 	/* See if any machine dependent calls. In theory, we would want
826 	 * to call the CPU first, and call the ppc_md. one if the CPU
827 	 * one returns a positive number. However there is existing code
828 	 * that assumes the board gets a first chance, so let's keep it
829 	 * that way for now and fix things later. --BenH.
830 	 */
831 	if (ppc_md.machine_check_exception)
832 		recover = ppc_md.machine_check_exception(regs);
833 	else if (cur_cpu_spec->machine_check)
834 		recover = cur_cpu_spec->machine_check(regs);
835 
836 	if (recover > 0)
837 		goto bail;
838 
839 	if (debugger_fault_handler(regs))
840 		goto bail;
841 
842 	if (check_io_access(regs))
843 		goto bail;
844 
845 	die_mce("Machine check", regs, SIGBUS);
846 
847 bail:
848 	/* Must die if the interrupt is not recoverable */
849 	if (regs_is_unrecoverable(regs))
850 		die_mce("Unrecoverable Machine check", regs, SIGBUS);
851 }
852 
853 #ifdef CONFIG_PPC_BOOK3S_64
854 DEFINE_INTERRUPT_HANDLER_RAW(machine_check_early_boot)
855 {
856 	udbg_printf("Machine check (early boot)\n");
857 	udbg_printf("SRR0=0x%016lx   SRR1=0x%016lx\n", regs->nip, regs->msr);
858 	udbg_printf(" DAR=0x%016lx  DSISR=0x%08lx\n", regs->dar, regs->dsisr);
859 	udbg_printf("  LR=0x%016lx     R1=0x%08lx\n", regs->link, regs->gpr[1]);
860 	udbg_printf("------\n");
861 	die("Machine check (early boot)", regs, SIGBUS);
862 	for (;;)
863 		;
864 	return 0;
865 }
866 
867 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception_async)
868 {
869 	__machine_check_exception(regs);
870 }
871 #endif
872 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception)
873 {
874 	__machine_check_exception(regs);
875 
876 	return 0;
877 }
878 
879 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */
880 {
881 	die("System Management Interrupt", regs, SIGABRT);
882 }
883 
884 #ifdef CONFIG_VSX
885 static void p9_hmi_special_emu(struct pt_regs *regs)
886 {
887 	unsigned int ra, rb, t, i, sel, instr, rc;
888 	const void __user *addr;
889 	u8 vbuf[16] __aligned(16), *vdst;
890 	unsigned long ea, msr, msr_mask;
891 	bool swap;
892 
893 	if (__get_user(instr, (unsigned int __user *)regs->nip))
894 		return;
895 
896 	/*
897 	 * lxvb16x	opcode: 0x7c0006d8
898 	 * lxvd2x	opcode: 0x7c000698
899 	 * lxvh8x	opcode: 0x7c000658
900 	 * lxvw4x	opcode: 0x7c000618
901 	 */
902 	if ((instr & 0xfc00073e) != 0x7c000618) {
903 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
904 			 " instr=%08x\n",
905 			 smp_processor_id(), current->comm, current->pid,
906 			 regs->nip, instr);
907 		return;
908 	}
909 
910 	/* Grab vector registers into the task struct */
911 	msr = regs->msr; /* Grab msr before we flush the bits */
912 	flush_vsx_to_thread(current);
913 	enable_kernel_altivec();
914 
915 	/*
916 	 * Is userspace running with a different endian (this is rare but
917 	 * not impossible)
918 	 */
919 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
920 
921 	/* Decode the instruction */
922 	ra = (instr >> 16) & 0x1f;
923 	rb = (instr >> 11) & 0x1f;
924 	t = (instr >> 21) & 0x1f;
925 	if (instr & 1)
926 		vdst = (u8 *)&current->thread.vr_state.vr[t];
927 	else
928 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
929 
930 	/* Grab the vector address */
931 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
932 	if (is_32bit_task())
933 		ea &= 0xfffffffful;
934 	addr = (__force const void __user *)ea;
935 
936 	/* Check it */
937 	if (!access_ok(addr, 16)) {
938 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
939 			 " instr=%08x addr=%016lx\n",
940 			 smp_processor_id(), current->comm, current->pid,
941 			 regs->nip, instr, (unsigned long)addr);
942 		return;
943 	}
944 
945 	/* Read the vector */
946 	rc = 0;
947 	if ((unsigned long)addr & 0xfUL)
948 		/* unaligned case */
949 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
950 	else
951 		__get_user_atomic_128_aligned(vbuf, addr, rc);
952 	if (rc) {
953 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
954 			 " instr=%08x addr=%016lx\n",
955 			 smp_processor_id(), current->comm, current->pid,
956 			 regs->nip, instr, (unsigned long)addr);
957 		return;
958 	}
959 
960 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
961 		 " instr=%08x addr=%016lx\n",
962 		 smp_processor_id(), current->comm, current->pid, regs->nip,
963 		 instr, (unsigned long) addr);
964 
965 	/* Grab instruction "selector" */
966 	sel = (instr >> 6) & 3;
967 
968 	/*
969 	 * Check to make sure the facility is actually enabled. This
970 	 * could happen if we get a false positive hit.
971 	 *
972 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
973 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
974 	 */
975 	msr_mask = MSR_VSX;
976 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
977 		msr_mask = MSR_VEC;
978 	if (!(msr & msr_mask)) {
979 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
980 			 " instr=%08x msr:%016lx\n",
981 			 smp_processor_id(), current->comm, current->pid,
982 			 regs->nip, instr, msr);
983 		return;
984 	}
985 
986 	/* Do logging here before we modify sel based on endian */
987 	switch (sel) {
988 	case 0:	/* lxvw4x */
989 		PPC_WARN_EMULATED(lxvw4x, regs);
990 		break;
991 	case 1: /* lxvh8x */
992 		PPC_WARN_EMULATED(lxvh8x, regs);
993 		break;
994 	case 2: /* lxvd2x */
995 		PPC_WARN_EMULATED(lxvd2x, regs);
996 		break;
997 	case 3: /* lxvb16x */
998 		PPC_WARN_EMULATED(lxvb16x, regs);
999 		break;
1000 	}
1001 
1002 #ifdef __LITTLE_ENDIAN__
1003 	/*
1004 	 * An LE kernel stores the vector in the task struct as an LE
1005 	 * byte array (effectively swapping both the components and
1006 	 * the content of the components). Those instructions expect
1007 	 * the components to remain in ascending address order, so we
1008 	 * swap them back.
1009 	 *
1010 	 * If we are running a BE user space, the expectation is that
1011 	 * of a simple memcpy, so forcing the emulation to look like
1012 	 * a lxvb16x should do the trick.
1013 	 */
1014 	if (swap)
1015 		sel = 3;
1016 
1017 	switch (sel) {
1018 	case 0:	/* lxvw4x */
1019 		for (i = 0; i < 4; i++)
1020 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1021 		break;
1022 	case 1: /* lxvh8x */
1023 		for (i = 0; i < 8; i++)
1024 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1025 		break;
1026 	case 2: /* lxvd2x */
1027 		for (i = 0; i < 2; i++)
1028 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1029 		break;
1030 	case 3: /* lxvb16x */
1031 		for (i = 0; i < 16; i++)
1032 			vdst[i] = vbuf[15-i];
1033 		break;
1034 	}
1035 #else /* __LITTLE_ENDIAN__ */
1036 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1037 	if (!swap)
1038 		sel = 3;
1039 
1040 	/* Otherwise, we need to swap the content of the components */
1041 	switch (sel) {
1042 	case 0:	/* lxvw4x */
1043 		for (i = 0; i < 4; i++)
1044 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1045 		break;
1046 	case 1: /* lxvh8x */
1047 		for (i = 0; i < 8; i++)
1048 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1049 		break;
1050 	case 2: /* lxvd2x */
1051 		for (i = 0; i < 2; i++)
1052 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1053 		break;
1054 	case 3: /* lxvb16x */
1055 		memcpy(vdst, vbuf, 16);
1056 		break;
1057 	}
1058 #endif /* !__LITTLE_ENDIAN__ */
1059 
1060 	/* Go to next instruction */
1061 	regs_add_return_ip(regs, 4);
1062 }
1063 #endif /* CONFIG_VSX */
1064 
1065 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception)
1066 {
1067 	struct pt_regs *old_regs;
1068 
1069 	old_regs = set_irq_regs(regs);
1070 
1071 #ifdef CONFIG_VSX
1072 	/* Real mode flagged P9 special emu is needed */
1073 	if (local_paca->hmi_p9_special_emu) {
1074 		local_paca->hmi_p9_special_emu = 0;
1075 
1076 		/*
1077 		 * We don't want to take page faults while doing the
1078 		 * emulation, we just replay the instruction if necessary.
1079 		 */
1080 		pagefault_disable();
1081 		p9_hmi_special_emu(regs);
1082 		pagefault_enable();
1083 	}
1084 #endif /* CONFIG_VSX */
1085 
1086 	if (ppc_md.handle_hmi_exception)
1087 		ppc_md.handle_hmi_exception(regs);
1088 
1089 	set_irq_regs(old_regs);
1090 }
1091 
1092 DEFINE_INTERRUPT_HANDLER(unknown_exception)
1093 {
1094 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1095 	       regs->nip, regs->msr, regs->trap);
1096 
1097 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1098 }
1099 
1100 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception)
1101 {
1102 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1103 	       regs->nip, regs->msr, regs->trap);
1104 
1105 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1106 }
1107 
1108 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception)
1109 {
1110 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1111 	       regs->nip, regs->msr, regs->trap);
1112 
1113 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1114 
1115 	return 0;
1116 }
1117 
1118 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception)
1119 {
1120 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1121 					5, SIGTRAP) == NOTIFY_STOP)
1122 		return;
1123 	if (debugger_iabr_match(regs))
1124 		return;
1125 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1126 }
1127 
1128 DEFINE_INTERRUPT_HANDLER(RunModeException)
1129 {
1130 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1131 }
1132 
1133 static void __single_step_exception(struct pt_regs *regs)
1134 {
1135 	clear_single_step(regs);
1136 	clear_br_trace(regs);
1137 
1138 	if (kprobe_post_handler(regs))
1139 		return;
1140 
1141 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1142 					5, SIGTRAP) == NOTIFY_STOP)
1143 		return;
1144 	if (debugger_sstep(regs))
1145 		return;
1146 
1147 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1148 }
1149 
1150 DEFINE_INTERRUPT_HANDLER(single_step_exception)
1151 {
1152 	__single_step_exception(regs);
1153 }
1154 
1155 /*
1156  * After we have successfully emulated an instruction, we have to
1157  * check if the instruction was being single-stepped, and if so,
1158  * pretend we got a single-step exception.  This was pointed out
1159  * by Kumar Gala.  -- paulus
1160  */
1161 void emulate_single_step(struct pt_regs *regs)
1162 {
1163 	if (single_stepping(regs))
1164 		__single_step_exception(regs);
1165 }
1166 
1167 static inline int __parse_fpscr(unsigned long fpscr)
1168 {
1169 	int ret = FPE_FLTUNK;
1170 
1171 	/* Invalid operation */
1172 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1173 		ret = FPE_FLTINV;
1174 
1175 	/* Overflow */
1176 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1177 		ret = FPE_FLTOVF;
1178 
1179 	/* Underflow */
1180 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1181 		ret = FPE_FLTUND;
1182 
1183 	/* Divide by zero */
1184 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1185 		ret = FPE_FLTDIV;
1186 
1187 	/* Inexact result */
1188 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1189 		ret = FPE_FLTRES;
1190 
1191 	return ret;
1192 }
1193 
1194 static void parse_fpe(struct pt_regs *regs)
1195 {
1196 	int code = 0;
1197 
1198 	flush_fp_to_thread(current);
1199 
1200 #ifdef CONFIG_PPC_FPU_REGS
1201 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1202 #endif
1203 
1204 	_exception(SIGFPE, regs, code, regs->nip);
1205 }
1206 
1207 /*
1208  * Illegal instruction emulation support.  Originally written to
1209  * provide the PVR to user applications using the mfspr rd, PVR.
1210  * Return non-zero if we can't emulate, or -EFAULT if the associated
1211  * memory access caused an access fault.  Return zero on success.
1212  *
1213  * There are a couple of ways to do this, either "decode" the instruction
1214  * or directly match lots of bits.  In this case, matching lots of
1215  * bits is faster and easier.
1216  *
1217  */
1218 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1219 {
1220 	u8 rT = (instword >> 21) & 0x1f;
1221 	u8 rA = (instword >> 16) & 0x1f;
1222 	u8 NB_RB = (instword >> 11) & 0x1f;
1223 	u32 num_bytes;
1224 	unsigned long EA;
1225 	int pos = 0;
1226 
1227 	/* Early out if we are an invalid form of lswx */
1228 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1229 		if ((rT == rA) || (rT == NB_RB))
1230 			return -EINVAL;
1231 
1232 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1233 
1234 	switch (instword & PPC_INST_STRING_MASK) {
1235 		case PPC_INST_LSWX:
1236 		case PPC_INST_STSWX:
1237 			EA += NB_RB;
1238 			num_bytes = regs->xer & 0x7f;
1239 			break;
1240 		case PPC_INST_LSWI:
1241 		case PPC_INST_STSWI:
1242 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1243 			break;
1244 		default:
1245 			return -EINVAL;
1246 	}
1247 
1248 	while (num_bytes != 0)
1249 	{
1250 		u8 val;
1251 		u32 shift = 8 * (3 - (pos & 0x3));
1252 
1253 		/* if process is 32-bit, clear upper 32 bits of EA */
1254 		if ((regs->msr & MSR_64BIT) == 0)
1255 			EA &= 0xFFFFFFFF;
1256 
1257 		switch ((instword & PPC_INST_STRING_MASK)) {
1258 			case PPC_INST_LSWX:
1259 			case PPC_INST_LSWI:
1260 				if (get_user(val, (u8 __user *)EA))
1261 					return -EFAULT;
1262 				/* first time updating this reg,
1263 				 * zero it out */
1264 				if (pos == 0)
1265 					regs->gpr[rT] = 0;
1266 				regs->gpr[rT] |= val << shift;
1267 				break;
1268 			case PPC_INST_STSWI:
1269 			case PPC_INST_STSWX:
1270 				val = regs->gpr[rT] >> shift;
1271 				if (put_user(val, (u8 __user *)EA))
1272 					return -EFAULT;
1273 				break;
1274 		}
1275 		/* move EA to next address */
1276 		EA += 1;
1277 		num_bytes--;
1278 
1279 		/* manage our position within the register */
1280 		if (++pos == 4) {
1281 			pos = 0;
1282 			if (++rT == 32)
1283 				rT = 0;
1284 		}
1285 	}
1286 
1287 	return 0;
1288 }
1289 
1290 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1291 {
1292 	u32 ra,rs;
1293 	unsigned long tmp;
1294 
1295 	ra = (instword >> 16) & 0x1f;
1296 	rs = (instword >> 21) & 0x1f;
1297 
1298 	tmp = regs->gpr[rs];
1299 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1300 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1301 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1302 	regs->gpr[ra] = tmp;
1303 
1304 	return 0;
1305 }
1306 
1307 static int emulate_isel(struct pt_regs *regs, u32 instword)
1308 {
1309 	u8 rT = (instword >> 21) & 0x1f;
1310 	u8 rA = (instword >> 16) & 0x1f;
1311 	u8 rB = (instword >> 11) & 0x1f;
1312 	u8 BC = (instword >> 6) & 0x1f;
1313 	u8 bit;
1314 	unsigned long tmp;
1315 
1316 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1317 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1318 
1319 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1320 
1321 	return 0;
1322 }
1323 
1324 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1325 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1326 {
1327         /* If we're emulating a load/store in an active transaction, we cannot
1328          * emulate it as the kernel operates in transaction suspended context.
1329          * We need to abort the transaction.  This creates a persistent TM
1330          * abort so tell the user what caused it with a new code.
1331 	 */
1332 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1333 		tm_enable();
1334 		tm_abort(cause);
1335 		return true;
1336 	}
1337 	return false;
1338 }
1339 #else
1340 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1341 {
1342 	return false;
1343 }
1344 #endif
1345 
1346 static int emulate_instruction(struct pt_regs *regs)
1347 {
1348 	u32 instword;
1349 	u32 rd;
1350 
1351 	if (!user_mode(regs))
1352 		return -EINVAL;
1353 
1354 	if (get_user(instword, (u32 __user *)(regs->nip)))
1355 		return -EFAULT;
1356 
1357 	/* Emulate the mfspr rD, PVR. */
1358 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1359 		PPC_WARN_EMULATED(mfpvr, regs);
1360 		rd = (instword >> 21) & 0x1f;
1361 		regs->gpr[rd] = mfspr(SPRN_PVR);
1362 		return 0;
1363 	}
1364 
1365 	/* Emulating the dcba insn is just a no-op.  */
1366 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1367 		PPC_WARN_EMULATED(dcba, regs);
1368 		return 0;
1369 	}
1370 
1371 	/* Emulate the mcrxr insn.  */
1372 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1373 		int shift = (instword >> 21) & 0x1c;
1374 		unsigned long msk = 0xf0000000UL >> shift;
1375 
1376 		PPC_WARN_EMULATED(mcrxr, regs);
1377 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1378 		regs->xer &= ~0xf0000000UL;
1379 		return 0;
1380 	}
1381 
1382 	/* Emulate load/store string insn. */
1383 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1384 		if (tm_abort_check(regs,
1385 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1386 			return -EINVAL;
1387 		PPC_WARN_EMULATED(string, regs);
1388 		return emulate_string_inst(regs, instword);
1389 	}
1390 
1391 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1392 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1393 		PPC_WARN_EMULATED(popcntb, regs);
1394 		return emulate_popcntb_inst(regs, instword);
1395 	}
1396 
1397 	/* Emulate isel (Integer Select) instruction */
1398 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1399 		PPC_WARN_EMULATED(isel, regs);
1400 		return emulate_isel(regs, instword);
1401 	}
1402 
1403 	/* Emulate sync instruction variants */
1404 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1405 		PPC_WARN_EMULATED(sync, regs);
1406 		asm volatile("sync");
1407 		return 0;
1408 	}
1409 
1410 #ifdef CONFIG_PPC64
1411 	/* Emulate the mfspr rD, DSCR. */
1412 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1413 		PPC_INST_MFSPR_DSCR_USER) ||
1414 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1415 		PPC_INST_MFSPR_DSCR)) &&
1416 			cpu_has_feature(CPU_FTR_DSCR)) {
1417 		PPC_WARN_EMULATED(mfdscr, regs);
1418 		rd = (instword >> 21) & 0x1f;
1419 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1420 		return 0;
1421 	}
1422 	/* Emulate the mtspr DSCR, rD. */
1423 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1424 		PPC_INST_MTSPR_DSCR_USER) ||
1425 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1426 		PPC_INST_MTSPR_DSCR)) &&
1427 			cpu_has_feature(CPU_FTR_DSCR)) {
1428 		PPC_WARN_EMULATED(mtdscr, regs);
1429 		rd = (instword >> 21) & 0x1f;
1430 		current->thread.dscr = regs->gpr[rd];
1431 		current->thread.dscr_inherit = 1;
1432 		mtspr(SPRN_DSCR, current->thread.dscr);
1433 		return 0;
1434 	}
1435 #endif
1436 
1437 	return -EINVAL;
1438 }
1439 
1440 int is_valid_bugaddr(unsigned long addr)
1441 {
1442 	return is_kernel_addr(addr);
1443 }
1444 
1445 #ifdef CONFIG_MATH_EMULATION
1446 static int emulate_math(struct pt_regs *regs)
1447 {
1448 	int ret;
1449 
1450 	ret = do_mathemu(regs);
1451 	if (ret >= 0)
1452 		PPC_WARN_EMULATED(math, regs);
1453 
1454 	switch (ret) {
1455 	case 0:
1456 		emulate_single_step(regs);
1457 		return 0;
1458 	case 1: {
1459 			int code = 0;
1460 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1461 			_exception(SIGFPE, regs, code, regs->nip);
1462 			return 0;
1463 		}
1464 	case -EFAULT:
1465 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1466 		return 0;
1467 	}
1468 
1469 	return -1;
1470 }
1471 #else
1472 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1473 #endif
1474 
1475 static void do_program_check(struct pt_regs *regs)
1476 {
1477 	unsigned int reason = get_reason(regs);
1478 
1479 	/* We can now get here via a FP Unavailable exception if the core
1480 	 * has no FPU, in that case the reason flags will be 0 */
1481 
1482 	if (reason & REASON_FP) {
1483 		/* IEEE FP exception */
1484 		parse_fpe(regs);
1485 		return;
1486 	}
1487 	if (reason & REASON_TRAP) {
1488 		unsigned long bugaddr;
1489 		/* Debugger is first in line to stop recursive faults in
1490 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1491 		if (debugger_bpt(regs))
1492 			return;
1493 
1494 		if (kprobe_handler(regs))
1495 			return;
1496 
1497 		/* trap exception */
1498 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1499 				== NOTIFY_STOP)
1500 			return;
1501 
1502 		bugaddr = regs->nip;
1503 		/*
1504 		 * Fixup bugaddr for BUG_ON() in real mode
1505 		 */
1506 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1507 			bugaddr += PAGE_OFFSET;
1508 
1509 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1510 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1511 			regs_add_return_ip(regs, 4);
1512 			return;
1513 		}
1514 
1515 		/* User mode considers other cases after enabling IRQs */
1516 		if (!user_mode(regs)) {
1517 			_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1518 			return;
1519 		}
1520 	}
1521 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1522 	if (reason & REASON_TM) {
1523 		/* This is a TM "Bad Thing Exception" program check.
1524 		 * This occurs when:
1525 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1526 		 *    transition in TM states.
1527 		 * -  A trechkpt is attempted when transactional.
1528 		 * -  A treclaim is attempted when non transactional.
1529 		 * -  A tend is illegally attempted.
1530 		 * -  writing a TM SPR when transactional.
1531 		 *
1532 		 * If usermode caused this, it's done something illegal and
1533 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1534 		 * operand to distinguish from the instruction just being bad
1535 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1536 		 * illegal /placement/ of a valid instruction.
1537 		 */
1538 		if (user_mode(regs)) {
1539 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1540 			return;
1541 		} else {
1542 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1543 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1544 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1545 			die("Unrecoverable exception", regs, SIGABRT);
1546 		}
1547 	}
1548 #endif
1549 
1550 	/*
1551 	 * If we took the program check in the kernel skip down to sending a
1552 	 * SIGILL. The subsequent cases all relate to user space, such as
1553 	 * emulating instructions which we should only do for user space. We
1554 	 * also do not want to enable interrupts for kernel faults because that
1555 	 * might lead to further faults, and loose the context of the original
1556 	 * exception.
1557 	 */
1558 	if (!user_mode(regs))
1559 		goto sigill;
1560 
1561 	interrupt_cond_local_irq_enable(regs);
1562 
1563 	/*
1564 	 * (reason & REASON_TRAP) is mostly handled before enabling IRQs,
1565 	 * except get_user_instr() can sleep so we cannot reliably inspect the
1566 	 * current instruction in that context. Now that we know we are
1567 	 * handling a user space trap and can sleep, we can check if the trap
1568 	 * was a hashchk failure.
1569 	 */
1570 	if (reason & REASON_TRAP) {
1571 		if (cpu_has_feature(CPU_FTR_DEXCR_NPHIE)) {
1572 			ppc_inst_t insn;
1573 
1574 			if (get_user_instr(insn, (void __user *)regs->nip)) {
1575 				_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1576 				return;
1577 			}
1578 
1579 			if (ppc_inst_primary_opcode(insn) == 31 &&
1580 			    get_xop(ppc_inst_val(insn)) == OP_31_XOP_HASHCHK) {
1581 				_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1582 				return;
1583 			}
1584 		}
1585 
1586 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1587 		return;
1588 	}
1589 
1590 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1591 	 * but there seems to be a hardware bug on the 405GP (RevD)
1592 	 * that means ESR is sometimes set incorrectly - either to
1593 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1594 	 * hardware people - not sure if it can happen on any illegal
1595 	 * instruction or only on FP instructions, whether there is a
1596 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1597 	 */
1598 	if (!emulate_math(regs))
1599 		return;
1600 
1601 	/* Try to emulate it if we should. */
1602 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1603 		switch (emulate_instruction(regs)) {
1604 		case 0:
1605 			regs_add_return_ip(regs, 4);
1606 			emulate_single_step(regs);
1607 			return;
1608 		case -EFAULT:
1609 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1610 			return;
1611 		}
1612 	}
1613 
1614 sigill:
1615 	if (reason & REASON_PRIVILEGED)
1616 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1617 	else
1618 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1619 
1620 }
1621 
1622 DEFINE_INTERRUPT_HANDLER(program_check_exception)
1623 {
1624 	do_program_check(regs);
1625 }
1626 
1627 /*
1628  * This occurs when running in hypervisor mode on POWER6 or later
1629  * and an illegal instruction is encountered.
1630  */
1631 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt)
1632 {
1633 	regs_set_return_msr(regs, regs->msr | REASON_ILLEGAL);
1634 	do_program_check(regs);
1635 }
1636 
1637 DEFINE_INTERRUPT_HANDLER(alignment_exception)
1638 {
1639 	int sig, code, fixed = 0;
1640 	unsigned long  reason;
1641 
1642 	interrupt_cond_local_irq_enable(regs);
1643 
1644 	reason = get_reason(regs);
1645 	if (reason & REASON_BOUNDARY) {
1646 		sig = SIGBUS;
1647 		code = BUS_ADRALN;
1648 		goto bad;
1649 	}
1650 
1651 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1652 		return;
1653 
1654 	/* we don't implement logging of alignment exceptions */
1655 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1656 		fixed = fix_alignment(regs);
1657 
1658 	if (fixed == 1) {
1659 		/* skip over emulated instruction */
1660 		regs_add_return_ip(regs, inst_length(reason));
1661 		emulate_single_step(regs);
1662 		return;
1663 	}
1664 
1665 	/* Operand address was bad */
1666 	if (fixed == -EFAULT) {
1667 		sig = SIGSEGV;
1668 		code = SEGV_ACCERR;
1669 	} else {
1670 		sig = SIGBUS;
1671 		code = BUS_ADRALN;
1672 	}
1673 bad:
1674 	if (user_mode(regs))
1675 		_exception(sig, regs, code, regs->dar);
1676 	else
1677 		bad_page_fault(regs, sig);
1678 }
1679 
1680 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception)
1681 {
1682 	die("Kernel stack overflow", regs, SIGSEGV);
1683 }
1684 
1685 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception)
1686 {
1687 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1688 			  "%lx at %lx\n", regs->trap, regs->nip);
1689 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1690 }
1691 
1692 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception)
1693 {
1694 	if (user_mode(regs)) {
1695 		/* A user program has executed an altivec instruction,
1696 		   but this kernel doesn't support altivec. */
1697 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1698 		return;
1699 	}
1700 
1701 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1702 			"%lx at %lx\n", regs->trap, regs->nip);
1703 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1704 }
1705 
1706 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception)
1707 {
1708 	if (user_mode(regs)) {
1709 		/* A user program has executed an vsx instruction,
1710 		   but this kernel doesn't support vsx. */
1711 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1712 		return;
1713 	}
1714 
1715 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1716 			"%lx at %lx\n", regs->trap, regs->nip);
1717 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1718 }
1719 
1720 #ifdef CONFIG_PPC_BOOK3S_64
1721 static void tm_unavailable(struct pt_regs *regs)
1722 {
1723 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1724 	if (user_mode(regs)) {
1725 		current->thread.load_tm++;
1726 		regs_set_return_msr(regs, regs->msr | MSR_TM);
1727 		tm_enable();
1728 		tm_restore_sprs(&current->thread);
1729 		return;
1730 	}
1731 #endif
1732 	pr_emerg("Unrecoverable TM Unavailable Exception "
1733 			"%lx at %lx\n", regs->trap, regs->nip);
1734 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1735 }
1736 
1737 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception)
1738 {
1739 	static char *facility_strings[] = {
1740 		[FSCR_FP_LG] = "FPU",
1741 		[FSCR_VECVSX_LG] = "VMX/VSX",
1742 		[FSCR_DSCR_LG] = "DSCR",
1743 		[FSCR_PM_LG] = "PMU SPRs",
1744 		[FSCR_BHRB_LG] = "BHRB",
1745 		[FSCR_TM_LG] = "TM",
1746 		[FSCR_EBB_LG] = "EBB",
1747 		[FSCR_TAR_LG] = "TAR",
1748 		[FSCR_MSGP_LG] = "MSGP",
1749 		[FSCR_SCV_LG] = "SCV",
1750 		[FSCR_PREFIX_LG] = "PREFIX",
1751 	};
1752 	char *facility = "unknown";
1753 	u64 value;
1754 	u32 instword, rd;
1755 	u8 status;
1756 	bool hv;
1757 
1758 	hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL);
1759 	if (hv)
1760 		value = mfspr(SPRN_HFSCR);
1761 	else
1762 		value = mfspr(SPRN_FSCR);
1763 
1764 	status = value >> 56;
1765 	if ((hv || status >= 2) &&
1766 	    (status < ARRAY_SIZE(facility_strings)) &&
1767 	    facility_strings[status])
1768 		facility = facility_strings[status];
1769 
1770 	/* We should not have taken this interrupt in kernel */
1771 	if (!user_mode(regs)) {
1772 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1773 			 facility, status, regs->nip);
1774 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1775 	}
1776 
1777 	interrupt_cond_local_irq_enable(regs);
1778 
1779 	if (status == FSCR_DSCR_LG) {
1780 		/*
1781 		 * User is accessing the DSCR register using the problem
1782 		 * state only SPR number (0x03) either through a mfspr or
1783 		 * a mtspr instruction. If it is a write attempt through
1784 		 * a mtspr, then we set the inherit bit. This also allows
1785 		 * the user to write or read the register directly in the
1786 		 * future by setting via the FSCR DSCR bit. But in case it
1787 		 * is a read DSCR attempt through a mfspr instruction, we
1788 		 * just emulate the instruction instead. This code path will
1789 		 * always emulate all the mfspr instructions till the user
1790 		 * has attempted at least one mtspr instruction. This way it
1791 		 * preserves the same behaviour when the user is accessing
1792 		 * the DSCR through privilege level only SPR number (0x11)
1793 		 * which is emulated through illegal instruction exception.
1794 		 * We always leave HFSCR DSCR set.
1795 		 */
1796 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1797 			pr_err("Failed to fetch the user instruction\n");
1798 			return;
1799 		}
1800 
1801 		/* Write into DSCR (mtspr 0x03, RS) */
1802 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1803 				== PPC_INST_MTSPR_DSCR_USER) {
1804 			rd = (instword >> 21) & 0x1f;
1805 			current->thread.dscr = regs->gpr[rd];
1806 			current->thread.dscr_inherit = 1;
1807 			current->thread.fscr |= FSCR_DSCR;
1808 			mtspr(SPRN_FSCR, current->thread.fscr);
1809 		}
1810 
1811 		/* Read from DSCR (mfspr RT, 0x03) */
1812 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1813 				== PPC_INST_MFSPR_DSCR_USER) {
1814 			if (emulate_instruction(regs)) {
1815 				pr_err("DSCR based mfspr emulation failed\n");
1816 				return;
1817 			}
1818 			regs_add_return_ip(regs, 4);
1819 			emulate_single_step(regs);
1820 		}
1821 		return;
1822 	}
1823 
1824 	if (status == FSCR_TM_LG) {
1825 		/*
1826 		 * If we're here then the hardware is TM aware because it
1827 		 * generated an exception with FSRM_TM set.
1828 		 *
1829 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1830 		 * told us not to do TM, or the kernel is not built with TM
1831 		 * support.
1832 		 *
1833 		 * If both of those things are true, then userspace can spam the
1834 		 * console by triggering the printk() below just by continually
1835 		 * doing tbegin (or any TM instruction). So in that case just
1836 		 * send the process a SIGILL immediately.
1837 		 */
1838 		if (!cpu_has_feature(CPU_FTR_TM))
1839 			goto out;
1840 
1841 		tm_unavailable(regs);
1842 		return;
1843 	}
1844 
1845 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1846 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1847 
1848 out:
1849 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1850 }
1851 #endif
1852 
1853 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1854 
1855 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm)
1856 {
1857 	/* Note:  This does not handle any kind of FP laziness. */
1858 
1859 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1860 		 regs->nip, regs->msr);
1861 
1862         /* We can only have got here if the task started using FP after
1863          * beginning the transaction.  So, the transactional regs are just a
1864          * copy of the checkpointed ones.  But, we still need to recheckpoint
1865          * as we're enabling FP for the process; it will return, abort the
1866          * transaction, and probably retry but now with FP enabled.  So the
1867          * checkpointed FP registers need to be loaded.
1868 	 */
1869 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1870 
1871 	/*
1872 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1873 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1874 	 *
1875 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1876 	 * recheckpoint.
1877 	 */
1878 
1879 	/* Enable FP for the task: */
1880 	current->thread.load_fp = 1;
1881 
1882 	/*
1883 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1884 	 */
1885 	tm_recheckpoint(&current->thread);
1886 }
1887 
1888 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm)
1889 {
1890 	/* See the comments in fp_unavailable_tm().  This function operates
1891 	 * the same way.
1892 	 */
1893 
1894 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1895 		 "MSR=%lx\n",
1896 		 regs->nip, regs->msr);
1897 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1898 	current->thread.load_vec = 1;
1899 	tm_recheckpoint(&current->thread);
1900 	current->thread.used_vr = 1;
1901 }
1902 
1903 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm)
1904 {
1905 	/* See the comments in fp_unavailable_tm().  This works similarly,
1906 	 * though we're loading both FP and VEC registers in here.
1907 	 *
1908 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1909 	 * regs.  Either way, set MSR_VSX.
1910 	 */
1911 
1912 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1913 		 "MSR=%lx\n",
1914 		 regs->nip, regs->msr);
1915 
1916 	current->thread.used_vsr = 1;
1917 
1918 	/* This reclaims FP and/or VR regs if they're already enabled */
1919 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1920 
1921 	current->thread.load_vec = 1;
1922 	current->thread.load_fp = 1;
1923 
1924 	tm_recheckpoint(&current->thread);
1925 }
1926 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1927 
1928 #ifdef CONFIG_PPC64
1929 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi);
1930 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi)
1931 {
1932 	__this_cpu_inc(irq_stat.pmu_irqs);
1933 
1934 	perf_irq(regs);
1935 
1936 	return 0;
1937 }
1938 #endif
1939 
1940 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async);
1941 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async)
1942 {
1943 	__this_cpu_inc(irq_stat.pmu_irqs);
1944 
1945 	perf_irq(regs);
1946 }
1947 
1948 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception)
1949 {
1950 	/*
1951 	 * On 64-bit, if perf interrupts hit in a local_irq_disable
1952 	 * (soft-masked) region, we consider them as NMIs. This is required to
1953 	 * prevent hash faults on user addresses when reading callchains (and
1954 	 * looks better from an irq tracing perspective).
1955 	 */
1956 	if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs)))
1957 		performance_monitor_exception_nmi(regs);
1958 	else
1959 		performance_monitor_exception_async(regs);
1960 
1961 	return 0;
1962 }
1963 
1964 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1965 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1966 {
1967 	int changed = 0;
1968 	/*
1969 	 * Determine the cause of the debug event, clear the
1970 	 * event flags and send a trap to the handler. Torez
1971 	 */
1972 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1973 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1974 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1975 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1976 #endif
1977 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1978 			     5);
1979 		changed |= 0x01;
1980 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1981 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1982 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1983 			     6);
1984 		changed |= 0x01;
1985 	}  else if (debug_status & DBSR_IAC1) {
1986 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1987 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1988 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1989 			     1);
1990 		changed |= 0x01;
1991 	}  else if (debug_status & DBSR_IAC2) {
1992 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1993 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1994 			     2);
1995 		changed |= 0x01;
1996 	}  else if (debug_status & DBSR_IAC3) {
1997 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1998 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1999 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
2000 			     3);
2001 		changed |= 0x01;
2002 	}  else if (debug_status & DBSR_IAC4) {
2003 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
2004 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
2005 			     4);
2006 		changed |= 0x01;
2007 	}
2008 	/*
2009 	 * At the point this routine was called, the MSR(DE) was turned off.
2010 	 * Check all other debug flags and see if that bit needs to be turned
2011 	 * back on or not.
2012 	 */
2013 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2014 			       current->thread.debug.dbcr1))
2015 		regs_set_return_msr(regs, regs->msr | MSR_DE);
2016 	else
2017 		/* Make sure the IDM flag is off */
2018 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2019 
2020 	if (changed & 0x01)
2021 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
2022 }
2023 
2024 DEFINE_INTERRUPT_HANDLER(DebugException)
2025 {
2026 	unsigned long debug_status = regs->dsisr;
2027 
2028 	current->thread.debug.dbsr = debug_status;
2029 
2030 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
2031 	 * on server, it stops on the target of the branch. In order to simulate
2032 	 * the server behaviour, we thus restart right away with a single step
2033 	 * instead of stopping here when hitting a BT
2034 	 */
2035 	if (debug_status & DBSR_BT) {
2036 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
2037 
2038 		/* Disable BT */
2039 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
2040 		/* Clear the BT event */
2041 		mtspr(SPRN_DBSR, DBSR_BT);
2042 
2043 		/* Do the single step trick only when coming from userspace */
2044 		if (user_mode(regs)) {
2045 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
2046 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
2047 			regs_set_return_msr(regs, regs->msr | MSR_DE);
2048 			return;
2049 		}
2050 
2051 		if (kprobe_post_handler(regs))
2052 			return;
2053 
2054 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
2055 			       5, SIGTRAP) == NOTIFY_STOP) {
2056 			return;
2057 		}
2058 		if (debugger_sstep(regs))
2059 			return;
2060 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
2061 		regs_set_return_msr(regs, regs->msr & ~MSR_DE);
2062 
2063 		/* Disable instruction completion */
2064 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
2065 		/* Clear the instruction completion event */
2066 		mtspr(SPRN_DBSR, DBSR_IC);
2067 
2068 		if (kprobe_post_handler(regs))
2069 			return;
2070 
2071 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2072 			       5, SIGTRAP) == NOTIFY_STOP) {
2073 			return;
2074 		}
2075 
2076 		if (debugger_sstep(regs))
2077 			return;
2078 
2079 		if (user_mode(regs)) {
2080 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2081 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2082 					       current->thread.debug.dbcr1))
2083 				regs_set_return_msr(regs, regs->msr | MSR_DE);
2084 			else
2085 				/* Make sure the IDM bit is off */
2086 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2087 		}
2088 
2089 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2090 	} else
2091 		handle_debug(regs, debug_status);
2092 }
2093 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2094 
2095 #ifdef CONFIG_ALTIVEC
2096 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception)
2097 {
2098 	int err;
2099 
2100 	if (!user_mode(regs)) {
2101 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2102 		       " at %lx\n", regs->nip);
2103 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2104 	}
2105 
2106 	flush_altivec_to_thread(current);
2107 
2108 	PPC_WARN_EMULATED(altivec, regs);
2109 	err = emulate_altivec(regs);
2110 	if (err == 0) {
2111 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2112 		emulate_single_step(regs);
2113 		return;
2114 	}
2115 
2116 	if (err == -EFAULT) {
2117 		/* got an error reading the instruction */
2118 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2119 	} else {
2120 		/* didn't recognize the instruction */
2121 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2122 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2123 				   "in %s at %lx\n", current->comm, regs->nip);
2124 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2125 	}
2126 }
2127 #endif /* CONFIG_ALTIVEC */
2128 
2129 #ifdef CONFIG_PPC_85xx
2130 DEFINE_INTERRUPT_HANDLER(CacheLockingException)
2131 {
2132 	unsigned long error_code = regs->dsisr;
2133 
2134 	/* We treat cache locking instructions from the user
2135 	 * as priv ops, in the future we could try to do
2136 	 * something smarter
2137 	 */
2138 	if (error_code & (ESR_DLK|ESR_ILK))
2139 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2140 	return;
2141 }
2142 #endif /* CONFIG_PPC_85xx */
2143 
2144 #ifdef CONFIG_SPE
2145 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException)
2146 {
2147 	unsigned long spefscr;
2148 	int fpexc_mode;
2149 	int code = FPE_FLTUNK;
2150 	int err;
2151 
2152 	interrupt_cond_local_irq_enable(regs);
2153 
2154 	flush_spe_to_thread(current);
2155 
2156 	spefscr = current->thread.spefscr;
2157 	fpexc_mode = current->thread.fpexc_mode;
2158 
2159 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2160 		code = FPE_FLTOVF;
2161 	}
2162 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2163 		code = FPE_FLTUND;
2164 	}
2165 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2166 		code = FPE_FLTDIV;
2167 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2168 		code = FPE_FLTINV;
2169 	}
2170 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2171 		code = FPE_FLTRES;
2172 
2173 	err = do_spe_mathemu(regs);
2174 	if (err == 0) {
2175 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2176 		emulate_single_step(regs);
2177 		return;
2178 	}
2179 
2180 	if (err == -EFAULT) {
2181 		/* got an error reading the instruction */
2182 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2183 	} else if (err == -EINVAL) {
2184 		/* didn't recognize the instruction */
2185 		printk(KERN_ERR "unrecognized spe instruction "
2186 		       "in %s at %lx\n", current->comm, regs->nip);
2187 	} else {
2188 		_exception(SIGFPE, regs, code, regs->nip);
2189 	}
2190 
2191 	return;
2192 }
2193 
2194 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException)
2195 {
2196 	int err;
2197 
2198 	interrupt_cond_local_irq_enable(regs);
2199 
2200 	preempt_disable();
2201 	if (regs->msr & MSR_SPE)
2202 		giveup_spe(current);
2203 	preempt_enable();
2204 
2205 	regs_add_return_ip(regs, -4);
2206 	err = speround_handler(regs);
2207 	if (err == 0) {
2208 		regs_add_return_ip(regs, 4); /* skip emulated instruction */
2209 		emulate_single_step(regs);
2210 		return;
2211 	}
2212 
2213 	if (err == -EFAULT) {
2214 		/* got an error reading the instruction */
2215 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2216 	} else if (err == -EINVAL) {
2217 		/* didn't recognize the instruction */
2218 		printk(KERN_ERR "unrecognized spe instruction "
2219 		       "in %s at %lx\n", current->comm, regs->nip);
2220 	} else {
2221 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2222 		return;
2223 	}
2224 }
2225 #endif
2226 
2227 /*
2228  * We enter here if we get an unrecoverable exception, that is, one
2229  * that happened at a point where the RI (recoverable interrupt) bit
2230  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2231  * we therefore lost state by taking this exception.
2232  */
2233 void __noreturn unrecoverable_exception(struct pt_regs *regs)
2234 {
2235 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2236 		 regs->trap, regs->nip, regs->msr);
2237 	die("Unrecoverable exception", regs, SIGABRT);
2238 	/* die() should not return */
2239 	for (;;)
2240 		;
2241 }
2242 
2243 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2244 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException)
2245 {
2246 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2247 	mtspr(SPRN_TCR, mfspr(SPRN_TCR) & ~TCR_WIE);
2248 	return 0;
2249 }
2250 #endif
2251 
2252 /*
2253  * We enter here if we discover during exception entry that we are
2254  * running in supervisor mode with a userspace value in the stack pointer.
2255  */
2256 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack)
2257 {
2258 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2259 	       regs->gpr[1], regs->nip);
2260 	die("Bad kernel stack pointer", regs, SIGABRT);
2261 }
2262 
2263 #ifdef CONFIG_PPC_EMULATED_STATS
2264 
2265 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2266 
2267 struct ppc_emulated ppc_emulated = {
2268 #ifdef CONFIG_ALTIVEC
2269 	WARN_EMULATED_SETUP(altivec),
2270 #endif
2271 	WARN_EMULATED_SETUP(dcba),
2272 	WARN_EMULATED_SETUP(dcbz),
2273 	WARN_EMULATED_SETUP(fp_pair),
2274 	WARN_EMULATED_SETUP(isel),
2275 	WARN_EMULATED_SETUP(mcrxr),
2276 	WARN_EMULATED_SETUP(mfpvr),
2277 	WARN_EMULATED_SETUP(multiple),
2278 	WARN_EMULATED_SETUP(popcntb),
2279 	WARN_EMULATED_SETUP(spe),
2280 	WARN_EMULATED_SETUP(string),
2281 	WARN_EMULATED_SETUP(sync),
2282 	WARN_EMULATED_SETUP(unaligned),
2283 #ifdef CONFIG_MATH_EMULATION
2284 	WARN_EMULATED_SETUP(math),
2285 #endif
2286 #ifdef CONFIG_VSX
2287 	WARN_EMULATED_SETUP(vsx),
2288 #endif
2289 #ifdef CONFIG_PPC64
2290 	WARN_EMULATED_SETUP(mfdscr),
2291 	WARN_EMULATED_SETUP(mtdscr),
2292 	WARN_EMULATED_SETUP(lq_stq),
2293 	WARN_EMULATED_SETUP(lxvw4x),
2294 	WARN_EMULATED_SETUP(lxvh8x),
2295 	WARN_EMULATED_SETUP(lxvd2x),
2296 	WARN_EMULATED_SETUP(lxvb16x),
2297 #endif
2298 };
2299 
2300 u32 ppc_warn_emulated;
2301 
2302 void ppc_warn_emulated_print(const char *type)
2303 {
2304 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2305 			    type);
2306 }
2307 
2308 static int __init ppc_warn_emulated_init(void)
2309 {
2310 	struct dentry *dir;
2311 	unsigned int i;
2312 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2313 
2314 	dir = debugfs_create_dir("emulated_instructions",
2315 				 arch_debugfs_dir);
2316 
2317 	debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated);
2318 
2319 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++)
2320 		debugfs_create_u32(entries[i].name, 0644, dir,
2321 				   (u32 *)&entries[i].val.counter);
2322 
2323 	return 0;
2324 }
2325 
2326 device_initcall(ppc_warn_emulated_init);
2327 
2328 #endif /* CONFIG_PPC_EMULATED_STATS */
2329